Penerapan Neural Style Transfer sebagai Alat Visualisasi Desain Fesyen Pakaian Atasan

Fathulloh, Akmal Sulthon (2025) Penerapan Neural Style Transfer sebagai Alat Visualisasi Desain Fesyen Pakaian Atasan. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211047-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Penerapan teknologi kecerdasan buatan dalam industri fesyen memberikan peluang inovatif dalam proses desain visual. Salah satu metode yang memiliki potensi besar adalah Neural Style Transfer (NST), yaitu teknik yang memungkinkan perpindahan gaya visual dari sebuah gambar ke gambar lain. Namun, penerapan NST secara langsung pada citra fesyen sering kali menghasilkan distorsi pada bagian non-pakaian, sehingga kurang ideal sebagai alat bantu visualisasi desain. Penelitian ini mengusulkan metode integrasi antara NST dengan teknik segmentasi citra menggunakan Mask R-CNN, yang memungkinkan transfer gaya hanya pada area pakaian atasan yang ditargetkan. Sistem dibangun menggunakan model Fast Arbitrary Style Transfer dari TensorFlow Hub serta model segmentasi yang dilatih ulang menggunakan dataset ModaNet. Proses stilisasi dilakukan secara selektif dengan menggabungkan hasil segmentasi dan gambar hasil NST. Evaluasi terhadap sistem dilakukan melalui metrik Average Precision (AP) pada segmentasi dan observasi kualitas visual hasil stilisasi. Hasilnya menunjukkan bahwa pendekatan ini berhasil meningkatkan fokus stilisasi, mempertahankan latar belakang dan wajah model, serta menghasilkan visualisasi desain yang lebih realistis dan aplikatif. Dengan demikian, sistem ini berpotensi menjadi alat bantu eksplorasi desain yang efisien bagi desainer fesyen.
==============================================================================================================================
The application of artificial intelligence in the fashion industry offers innovative opportunities in the visual design process. One promising technique is Neural Style Transfer (NST), which enables the transfer of visual styles from one image to another. However, applying NST directly to fashion images often causes distortions in non-clothing regions, making it less ideal as a design visualization tool. This study proposes an integrated approach that combines NST with image segmentation techniques using Mask R-CNN, allowing style transfer to be applied only to the targeted clothing area (tops). The system was built using the Fast Arbitrary Style Transfer model from TensorFlow Hub and a segmentation model fine-tuned with the ModaNet dataset. Stylization is selectively applied by combining the segmentation result with the NST output. The system was evaluated using the Average Precision (AP) metric for segmentation and visual quality assessments of the stylized output. The results show that this approach successfully enhances style transfer focus, preserves the model’s background and face, and produces more realistic and applicable design visualizations. Thus, the system has the potential to serve as an efficient design exploration tool for fashion designers.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Neural Style Transfer, Segmentasi Citra, Mask R-CNN, Desain Fesyen, Neural Style Transfer, Image Segmentation, Mask R-CNN, Fashion Design
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Information Technology > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Akmal Sulthon Fathulloh
Date Deposited:	30 Jul 2025 09:32
Last Modified:	30 Jul 2025 09:32
URI:	http://repository.its.ac.id/id/eprint/124369

Actions (login required)

View Item